Future neural networks may rely on artificial synapses, better for mimicking the human brain.
Computers are incredible calculating machines, and can do a lot of things, but they’ve never been good at replicating the low power processing capability of the human brain. Researchers from Stanford University and Sandia National Laboratories have now made an advancement which they hope can bridge the gap: They have created an artificial version of a synapse, the slight gap, and junction, between brain cells that allows them to communicate. Their results were published earlier this week in the journal Nature Materials.
“It works like a real synapse but it’s an organic electronic device that can be engineered,” said Alberto Salleo, associate professor of materials science and engineering at Stanford and senior author of the paper. “It’s an entirely new family of devices because this type of architecture has not been shown before. For many key metrics, it also performs better than anything that’s been done before with inorganics.”
The unique feature of the artificial synapse is that, like in the real brain, the synapses learn from the signals that cross them. This saves a lot of energy compared to a conventional computer which stores and process information separately. In the artificial synapse, processing the information and storing the information are one and the same.
There are many potential applications for the technology, one of which is neural networks, something we see more and more in AI systems. “Deep learning algorithms are very powerful but they rely on processors to calculate and simulate the electrical states and store them somewhere else, which is inefficient in terms of energy and time,” said Yoeri van de Burgt, lead author of the paper. “Instead of simulating a neural network, our work is trying to make a neural network.”
The synapse consists of two thin metal films with an electrolyte of salty water between them. In essence, it works like a transistor, with one terminal determining the flow of electricity between the other two. So far, the device has been put to the test in a handwriting analysis test. The synapse was tasked with recognizing handwritten numbers 0 through 9, and had a success rate of 93-97%. This is a promising result, as computers often have trouble interpreting visual input.
source: Nature Materials